Differences

This shows you the differences between two versions of the page.

--- courses:xai:p1_2024 [2024/11/22 07:51] – created admin
+++ courses:xai:p1_2024 [2024/11/22 12:35] (current) – [Explainable Actrive Learning] admin
@@ Line 3: / Line 3: @@
 Assignments marked with {{:courses:xai:winner.png?25|}} can be extended as a continution in Programming Assgnment II.
 Projects marked with {{:courses:xai:contract.png?25|}} can be (with some additional work and depending on the results) published as scientific papers.
+===== SHAP for time series  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+Main goal of the project is to take the code: https://github.com/sbobek/WindowSHAP/tree/main
+And add possibility of adding dynamic windows using ruptures point-detection algorithm.
+===== Study Rashomon Effect  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+Main goal of the project is to reproduce methods from: https://arxiv.org/pdf/2306.15786 or https://arxiv.org/pdf/2405.01557 And possibly extend with more exaples/models for more extensive evaluaiton.
+===== LUX Speedup  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+Speedup LUX explainer and confirm that after the speedup the results are not changed (Different wasy of speedup can be considered).
+===== PPCF Counterfactual explainer evaluation  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+The goal is to take: https://github.com/ofurman/counterfactuals/tree/main and evaluate it on more examples
+===== Shap on OPTUNA  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+The goal is to run hyperparameter optimization algorithm such as optuna, collect results from it and use SHAP to try to explain the impact of different hyperparameters on the quality of the optimization process
+===== Explainable Active Learning  =====
+{{:courses:xai:winner.png?30|}}
+{{:courses:xai:contract.png?30|}}
+The goal is to use SHAP gradients (implemented in LUX) to select instances that should be labelled by expert (Active Learning paradigm) and compare with random sampling.